Open source could pop the AI bubble — and soon - FT中文网
登录×
电子邮件/用户名
密码
记住我
请输入邮箱和密码进行绑定操作:
请输入手机号码,通过短信验证(目前仅支持中国大陆地区的手机号):
请您阅读我们的用户注册协议隐私权保护政策,点击下方按钮即视为您接受。
人工智能

Open source could pop the AI bubble — and soon

Valuations rest on the incorrect assumption that frontier model creators have built massive, durable moats
00:00

{"text":[[{"start":6.24,"text":"The writer is an open-internet activist and president of the Mozilla Foundation"}],[{"start":11.18,"text":"We cannot go a day without people asking: when will the artificial intelligence bubble pop and what will be the cause? I increasingly think that open-source AI models may be the pin that pops the bubble. "}],[{"start":25.91,"text":"Current AI valuations assume massive, durable moats. Investors have priced in the assumption that only a few companies can build frontier AI models, allowing them to extract monopoly rents. "}],[{"start":40.75,"text":"But if open-source models can match the performance of closed models at a fraction of the cost, that assumption collapses. Think about the billions of dollars in venture investment in OpenAI and Anthropic — as well as internal investment at Google. If the market realises that closed models may be good for cutting-edge research but unnecessary for most AI applications, valuations plummet."}],[{"start":71.18,"text":"A recent study by MIT economist Frank Nagle shows exactly how good open AI models have become. They are on average six times cheaper to use than equivalent closed models. And they are narrowing the performance gap within a few months of each new closed-model release. The speed of this catch-up cycle is accelerating."}],[{"start":95.26,"text":"If users were to choose the best observable AI model based on both price and performance they could save $20bn-$48bn a year.  "}],[{"start":106.26,"text":"Of course, both OpenAI and Google also developed open models — free to use and somewhat flexible. But these often feel like side projects. They are unlikely to win market share unless the companies put more work into them."}],[{"start":122.30000000000001,"text":"The companies that are leading in open-source AI are Chinese. Models by DeepSeek and Alibaba regularly excel in widely used AI benchmarks. A recent study by MIT and developer platform Hugging Face showed a steep decline in the market share of open models from Google, Meta and OpenAI, and a sharp increase in the use of models like DeepSeek and Qwen."}],[{"start":150.25,"text":"Western companies and research labs are hot on their heels. French start-up Mistral has released all of its new models as open source. Mistral Large 3 debuted at number two among open-source non-reasoning models on the LMArena leaderboard, within striking distance of its Chinese competitors. And Seattle-based lab Ai2 released a new set of its Olmo models as truly open — meaning every step of the development pipeline was exposed, from the underlying training data to the parameters that define the model’s output. The extra flexibility and utility this creates are very useful for developers building apps and services on top of AI models."}],[{"start":193.8,"text":"What’s interesting — and worrisome — is that governments are moving to invest huge amounts in closed-model AI. European countries — and those like Canada, where I come from — talk a great deal about tech sovereignty and using AI to drive economic growth. Yet they seem to be ignoring the fact that leaning into open source could get them to these goals faster and cheaper than backing large tech companies."}],[{"start":220.55,"text":"More consequentially, whether governments integrate open-source AI models into their plans could be a major driver in whether they are winners or losers in the AI race over the long run. "}],[{"start":234.13000000000002,"text":"The tipping point could come soon. When a major enterprise announces that it’s switching from a closed model to an open alternative — or when a breakout consumer app is built on a truly open model — investors will realise that the game has changed.  "}],[{"start":250.72000000000003,"text":"The next question is: who will be the winners when the bubble pops? If the dotcom crash is anything to go by it could be the open-source AI model creators. The free, open-source operating system Linux — on which 90 per cent of the public cloud runs today — proved resilient in the aftermath of that crash. The same pattern could be repeated today. The companies with lower valuations now could have significant upside tomorrow."}],[{"start":288.88,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1766445521_9780.mp3"}

版权声明:本文版权归FT中文网所有,未经允许任何单位或个人不得转载,复制或以任何其他方式使用本文全部或部分,侵权必究。

苏格兰“威士忌湖”再现

由于苏格兰酿酒厂库存过剩,一些酒厂被迫暂停或缩减生产规模,这引发了人们对乡村经济的担忧。

年度关键词:AI泡沫

硅谷与华尔街的“高阶祭司”们开始承认大型科技股估值的确有些过火。

FT社评:英国资产折价,外资收购的盛宴与隐忧

外资收购金额激增,既映照出资产的优质底色,也暴露出其仍在折价交易的现实。

特朗普任命美国格陵兰岛特使

杰夫•兰德里的角色凸显了美国总统控制北极岛屿的决心。

福特押注“未来卡车”的电动化如何导致195亿美元减记

F-150 Lightning的经济账算不拢,也折射出全行业对美国电动汽车普及速度的误判。

一周展望:金融市场会在最后关头上演“圣诞行情”吗?

由于担忧人工智能公司的巨额开支,今年的“圣诞行情”迟迟没有到来。
设置字号×
最小
较小
默认
较大
最大
分享×